Research And Implementation Of Real-Time Target Tracking System Based On FPGA

Target tracking is one of the research hotspots of computer vision.In scientific research and practical applications,the accuracy and real-time performance of the target tracking system are highly demanded.As video images tend to be high-definition and the complexity of tracking algorithms increases,the target tracking system implemented by the software is weak in processing speed.FPGAs have powerful parallel processing capabilities,and can be reconfigured,with high integration and low power consumption.They can pipeline a large amount of data,which is very suitable for hardware acceleration of visual target tracking algorithms.Therefore,it is of great research significance to implement a target tracking system with strong real-time,small size and low power consumption based on FPGA.The main purpose of this paper is to use Xilinx's new generation FPGA chip Zynq7020 SOC as the core processing platform,to improve and optimize the problems in CamShift target tracking algorithm,and to complete the research and implementation of real-time target tracking system based on FPGA.The main work of this paper is as follows:1.A visual target tracking algorithm combined with improved CamShift algorithm and Kalman filtering is proposed.According to theoretical analysis,the CamShift algorithm is difficult to cope with in the scenes of severe occlusion,background interference and fast moving,and the tracking accuracy drops sharply.In view of the above problems,we use a strategy of occlusion coefficient decision to improve the CamShift algorithm.In order to suppress that the search window is abnormal when the target is occluded or the target window is similar to the background color.The occlusion coefficient is calculated to adaptively adjust the size of the search window,and the Kalman filtering algorithm is used to predict the position information of the target.When the target moves quickly,expand the search scope to search for the target.At the same time,an algorithm for dynamically adjusting the shape of the tracking window is proposed,which greatly improves the applicability of the tracking algorithm in different scenarios.Finally,the test on the international target tracking standard test(OTB100)shows that compared with the CamShift tracking algorithm,the accuracy and robustness of the target tracking algorithm we proposed are greatly improved.2.A real-time target tracking system based on FPGA is implemented.The visual target tracking algorithm of improved CamShift algorithm and Kalman filtering is accelerated and optimized on the Zynq7020 SOC chip.The PS section configures the CMOS camera,AXI DMA,and AXI VDMA IP cores.Hardware design for image acquisition,display and target tracking algorithms is performed in the PL section.The Verilog hardware description language is used to build a complete video stream channel,and the captured video images are transmitted in two channels.The pipeline is optimized and the image and tracking window are displayed in real time.The CamShift tracking algorithm iterates the data in the search window.The same frame image needs to access the memory multiple times.Ours system uses the combination of DMA and interrupt to subtly read the data in the search window.The Kalman filter module uses two parallel one-dimensional Kalman filters to predict the target position and logically multiplex the operations of the floating-point numbers to ensure the accuracy of the algorithm while saving a lot of resources.The target tracking algorithm module is designed as an IP core based on the AXI_Stream protocol,and communicates with the PS through the AXI_HP high-performance data interface,which further improves the processing speed of the tracking algorithm.3.The function module is simulated by VIVADO2017.4 and MATLAB,and the tracking performance of the target tracking system in the real scene is tested on the evaluation board.The experimental results show that the target tracking system we designed has good real-time performance,high tracking accuracy,and good robustness for scenarios such as target occlusion,background interference and fast motion.